Prediction of Pancreatogenic Diabetes After Partial Resection of the Pancreas

Overview

The primary aim of the study is to identify clinical, laboratory, imaging, and pathohistological risk factors for the development of pancreatogenic diabetes after partial resection of the pancreas (PRP) and, based on these, to build a predictive model that would enable reliable and accurate identification of patients who will develop pancreatogenic diabetes. A review of the literature shows that there is currently no validated tool available in clinical practice for identifying patients at risk of this type of diabetes after PRP. Etiological differences between type 2 diabetes and pancreatogenic diabetes often lead to late diagnosis, limiting opportunities for timely intervention. The development of a comprehensive predictive model would contribute to a better understanding of the pathophysiology of pancreatogenic diabetes, enable the identification of patients at risk, facilitate tailored clinical monitoring, and allow the implementation of targeted preventive or therapeutic measures in the perioperative period. Based on a review of the literature and clinical observations, the investigators formulated the following hypotheses: H1: Clinical, laboratory, imaging, and pathohistological variables are independently associated with the development of pancreatogenic diabetes after partial pancreatic resection. H2: A predictive model that allows reliable and accurate identification of patients who will develop pancreatogenic diabetes can be build based on the identified variables (H1). H3: A comprehensive predictive model in the perioperative period allows more precise identification of patients who will develop pancreatogenic diabetes, and thus more effective risk stratification compared to the use of individual risk factors (variables). Based on the reviewed literature and our own clinical observations, this constitutes an original contribution to science. The investigators expect that the research will identify key clinical, laboratory, imaging, and pathohistological factors associated with the development of pancreatogenic diabetes after partial pancreatic resection. The development of a clinically useful predictive model that will allow individual risk assessment for the development of pancreatogenic diabetes in patients after partial pancreatic resection is anticipated. A similar comprehensive model has not yet been described in the accessible literature. The research will contribute to a better understanding of metabolic changes and will enable the identification of patients who will develop pancreatogenic diabetes after partial pancreatic resection. The results of the study could provide a basis for improved perioperative monitoring of these patients in the field of pancreatogenic diabetes.

Research Protocol: At the first preoperative examination in the Abdominal Surgery Outpatient Clinic of the University Medical Centre Ljubljana, inclusion and exclusion criteria for the study will be considered and the suitable patients will recieve oral and written explanation about the purpuse, course and nature of the study (day 0). After obtaining written consent from patients, the investigators will collect demographic data (gender, age), data from each participant's medical records (associated diseases, list of regular therapies, etc.) and clinical data on the underlying pancreatic disease (including abdominal CT findings). The anthropometric measurements (body weight and hight), vital sign measurements (blood pressure, pulse) and laboratory markers, that are routinely analyzed before the PRP, will be determined. Additionally, HbA1c, serum fasting insulin and C-peptide will be measured and an oral glucose tolerance test (OGTT) administered,according to a standardized procedure with a standard fasting period. The degree of insulin resistance (HOMA-IR and CGR) will be assessed. Blood analyses will be conducted at the Laboratory of the Clinical Institute of Clinical Chemistry and Biochemistry and the Clinic of Nuclear Medicine of the University Medical Centre Ljubljana. According to clinical indications, the participants will then be called for the PRP (e.g., distal pancreatectomy, pancreatoduodenectomy, or other form of partial resection), which will be carried out at the Clinical Department of Abdominal Surgery at the University Medical Center in Ljubljana. The pathological examination of the removed part of the pancreas will be undertaken as part of the routine postoperative analysis at the Institute of Pathology, Faculty of Medicine, University of Ljubljana. In the removed pancreatic tissue the proportion of β cells, fat, and fibrous tissue will be assessed during patohistological analysis. Three months after PRP, which coincides with the regular postoperative check-up with the abdominal surgeon, there will be a follow-up examination in Abdominal Surgery Outpatient Clinic of the University Medical Centre Ljubljana. The same data will be collected as before PRP and HOMA-IR and CGR recalculated. Likewise, patients will undergo a follow-up abdominal CT as part of routine clinical practice, where using a computer program, the volume of tissue removed, as well as the fat and fibrous tissue will be compared to the examination before the PRP. This time point after the PRP was chosen as clinically feasible and methodologically suitable for assessing early postoperative endocrine dysfunction of the pancreas, as it allows monitoring of metabolic changes after resection with minimal additional burden on the patients. The three-month follow-up also reduces the impact of long-term confounding factors, such as the progression of malignant disease, cachexia, prolonged systemic oncological treatment, and loss of patients from follow-up. When interpreting the results, the investigators will nevertheless take into account postoperative complications, the introduction of chemotherapy, the use of glucocorticoids, nutritional status, and other factors that can affect glucose metabolism. All the obtained data will be statistically analyzed, and a statistical predictive model for the occurrence of pancreatogenic diabetes after partial pancreatic resection will be constructed.To reduce the possibility of variability between different observers, histopathological analyses will be performed by one pathologist and CT analyses by one radiologist. Statistical Methods: All approaches and methods presented below are based on the TRIPOD+AI guidelines for statistical analysis and preparation of predictive models in the field of biomedicine. The sample size was estimated based on recommendations for the development of clinical predictive models. With the expected incidence of pancreatogenic diabetes, which according to PRP is approximately 20-30%, the investigators expect about 30-45 events. Such a number of events allows the inclusion of a limited number of predictors in a multivariable statistical model while taking into account the ratio of the number of events to the number of predictive variables. At the same time, the number of participants also depends on the number of partial pancreatic resections at the Clinical Department of Abdominal Surgery at the University Medical Center Ljubljana. It was also considered that some patients will drop out of the study (either due to the prognosis of the underlying disease or other factors), so the planned sample size will be increased by 10%. Therefore it is estimated that at least 150 patients has to be included in the study. The analytical workflow consists of three main components: data preprocessing, statistical model building, and interpretation of model results. In the initial phase of analysis, a systematic understanding of the data will be carried out, which includes a review of all collected clinical, demographic, and laboratory measurement variables. Numerical variables will be analyzed using basic descriptive statistical indicators (mean, standard deviation, median, interquartile range), with the normality of the distribution assessed using appropriate statistical tests and graphical methods. Categorical variables will be summarized with frequencies and percentages. Missing variable values will be addressed according to their pattern of occurrence. If the proportion of missing data is ≤ 10%, appropriate multivariate methods for imputing missing values will be used; otherwise, a complete case analysis will be performed. The pre-selection of measurable variables will be carried out based on a combination of clinical judgment, review of the existing literature, and statistical analysis. In the initial phase, the variables will be evaluated using univariate statistical methods in relation to the dependent variable. Variables that reflect clinical relevance or statistically significant or borderline significant association with the outcome will be included in the subsequent analysis.The core of the task involves the development and evaluation of a statistical model for predicting the value of the dependent variable . Supervised machine learning approaches will be used; test standard algorithms for pre-symbolic and symbolic learning, including logistic regression, support vector machines, random forests, and the XGBoost approach. In order to further reduce the risk of overfitting the predictive model and to improve its stability and interpretability, the number of predictors will be further limited using wrapper and embedded approaches. The built models will be evaluated (solution stability and generalizability) according to the standard 10-fold cross-validation scheme. All approaches will be implemented in accordance with TRIPOD guidelines. The importance of individual predictors and their association with clinical parameters will be explained using the SHAP (SHapley Additive exPlanations) approach. The quality of the models will be evaluated using standard metrics for balanced data, including discrimination, calibration parameters, and (clinical) decision curve analysis . All analyses will be implemented in the R programming language. The use of the predictive model does not pose any additional interventions or burdens for the participants, as it is based solely on a retrospective analysis of collected data.